Benchmark

[AlixANNERAUD]

Thank you for your awesome project. The benchmark published looks promising, but your scripting language looks like many others, how do you manage to make it faster than others (Lua for example) ?
And do you have other benchmarks comparison (for example : Berry-Lang, Toit or regular C/C++ compiled code) ?

[jingoro2112]

No interpreted language will ever come within a few orders of magnitude of compiled, If you can compile code (and I include JIT languages) then that will always be the fastest and preferable solution. Compilation is not possible for many use cases. JIT compilation is not allowed on consoles (xbox/ps) for example since there are strict hardware controls on execution space for security reasons, but many projects call for code written by users. In my example I want users to write their own lightning scripts. Trivial, but I need them to execute on embedded hardware and requiring installation and compilation of the entire tool stack for the chip (and the associated hardware required to reprogram it in the field?) not an option. So I needed an interpreter that could run on extremely limited hardware (oh, and 8 of them simultaneously, one for each channel) yet be fast enough to maintain 60 updates/sec with somewhat mathy/arbitrary code? Didn't exist so I made wrench. I chose c semantics so the language would be easy to learn and "just work" for the most programmers possible. The secret sauce is the compiler that breaks it down into byte code before execution, it's not directly interpreted. I observed it executing a wide-range of tasks and added several "super instructions" that do many tasks in a single byte-code cycle, for example: if ( x < y ) { } else { } Is a single instruction The code itself is implemented with jump-tables wherever possible and profiled heavily to knock down any unexpected tall poles. To paraphrase one my my heroes Michael Abrash: Until the wrench interpreter is a single nop there will always be more cycles to find, but I think I've done a pretty good job of optimizing for now. It's much faster than lua, squirrel and way faster than Python ... If I need more speed I'll come back to it, and if someone else wants to take a crack at it, well that's why I published it for everyone to wrench on. (see what i did there? :)

…

-Curt

On Thu, Apr 6, 2023 at 2:50 AM Alix ANNERAUD ***@***.***> wrote: Thank you for your awesome project. The benchmark published looks promising, but your scripting language looks like many others, how do you manage to make it faster than others (Lua for example) ? And do you have other benchmarks comparison (for example : Berry-Lang, Toit or regular C/C++ compiled code) ? — Reply to this email directly, view it on GitHub <#8>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AALIKAZPIQACZSCANYMQWM3W7ZRR7ANCNFSM6AAAAAAWU72AKE> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[AlixANNERAUD]

Okay, thank you for your answer. Indeed, it makes sense. In any case it is a great project with impressive efficiency. I did some tests and it turned out that on my laptop (i5 10th gen), wrench passed the benchmark in 17 seconds, against 30 seconds for berry and 47 for wren. However, I still have a question, do you intend to implement the classes?

[jingoro2112]

I'd love to answer that question but I'm afraid I don't understand it, what do you mean by "classes"?

…

On Fri, Apr 7, 2023 at 10:14 AM Alix ANNERAUD ***@***.***> wrote: Okay, thank you for your answer. Indeed, it makes sense. In any case it is a great project with impressive efficiency. I did some tests and it turned out that on my laptop (i5 10th gen), wrench passed the benchmark in 17 seconds, against 30 seconds for berry and 47 for wren. However, I still have a question, do you intend to implement the classes? — Reply to this email directly, view it on GitHub <#8 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AALIKA4KALDBU7FNEIJY3FTXAAOLVANCNFSM6AAAAAAWU72AKE> . You are receiving this because you commented.Message ID: ***@***.***>